Recently I've done some testing that made me questioning about the behaviour of the combat table close to critical strike cap.
Common belief is that hits in the table are only happening if nothing else happened, thus there could be situations where you would never see a hit, but only crit, dodges, parries, misses and glances.
It is also known that against a boss class target there is a critical strike loss of about 4.8%. It is not confirmed yet if this value is valid for every class or for every weapon combination and only rogue's combat table has been extensively tested.
The discussion is here

During my tests I tried a crit capped setup hitting from the front and swapping out all hit and exp rating. Swinging with 1 and 2 handers I found out that expertise seems to reduce both dodge and parry and there also seems to be a ~5% hit that pushes out additional crits. This happened across tests with different expertise and crit rating equipped.

I'm collecting the data in a shared document that you can see at:
http://spreadsheets.google.com/ccc?k...yWGJITkE&hl=en

First test was made with no hit, 4.25% expertise +1% WM (-1% dodge), 40.77 paperdoll crit, dualwielder 1h as arms

Test 1
Swings	10036
crits	3307	(32.95%)
misses	2666	(26.56%)
glances	2402	(23.93%)
parries	1010	(10.06%)
dodges	110	(1.096%)
hits	541	(5.39%)

At the start I was surprised that hit and dodges were about 6.5%, so I initially thought that expertise was converted into hits rather than getting back some crit. But then I made some more tests as fury with 0 hit and 0 exp with higher crit ratings.

Test 3 was made dualwielding 2handers, Test 4 dualwielding 2 fast 1handers. Even if Test 4 had 2.8% more crit and double sample, the crit rating seemed stable, meaning I was above the cap. However I was still seeing aorund 5% hits.

Test made with 44.58% crit, 0 hit, 0 exp, hitting from the front dualwielding 2handers

Test 3
Swings	5093
c	1243	(24.4%)
m	1349	(26.49%)
g	1173	(23.03%)
p	711	(13.96%)
d	354	(6.95%)
h	263	(5.16%)

Test made with 47.37% crit, 0 hit, 0 exp, hitting from the front dualwielding 1handers

Test 4
Swings	10601
c	2506	(23.6%)
m	2890	(27.26%)
g	2492	(23.5%)
p	1462	(13.79%)
d	692	(6.53%)
h	559	(5.28%)

Test made with 46.59% crit, 0 hit, 3.5% exp (118 rating no racial)

Test 5
Swings	5845
c	1807	(30.91%)
m	1652	(28.26%)
g	1312	(22.45%)
p	628	(10.74%)
d	165	(2.82%)
h	281	(4.81%)

Test made with 46.59% crit, 0 hit, 6.5% exp (179 rating + orc racial)

Test 6
Swings	3117
c	1138	(36.51%)
m	855	(27.43%)
g	740	(23.74%)
p	238	(7.64%)
d	0	(0%)
h	146	(4.68%)

From what I've seen so far, even considering the small samples of tests 5 and 6, the trend of parry seems to get lower with expertise as much as dodge, and regardless of the crit you have over the softcap, there seems to be a 5% hit always in place. I estimate the parry rating of the target dummy to be about 14%, and the first test was made with Wepon Mastery that only reduces dodges, thus parry should have been reduced by around ~4.25% while dodges by ~5.25% and this seems to appear from the results (even if with some uncertainty).

I believe this 5% is the crit depression working in a way so that you have a minimum number of hits that your crit cannot prevent. Estimating the correct value needs a lot more samples since it's not even sure if this crit depression is flat or works differently between classes or weapon combinations.

I'm still running some tests on the lower level especially because Rallik in the "Retesting hit table" post found out that warriors could have a different value.

It would be nice to see more tests at lower and higher ends. If these things will happen consistently even when hitting from behind, we could say that the white critical cap is around 71% effective crit once hit and exp capped.

I'll try to update the spreadsheet since the collection is quite time-consuming especially when I can't use fast weapons. Any help, suggestion or confutation is appreciated.

It looks to me like Blizz coded crit depression by forcing 4.8% (or whatever it is for a given class) of swings to always hit. Maybe there is a second type of "hit" that is above crits in the combat table. If WowWiki's listed combat table is correct (a big assumption, I know) then it would sit between Block and Crit since we know it's possible for tanks to become unhittable, at least against npcs.

That assumes that the NPC -> PC hit table is the same as the PC -> NPC table; we know that they are not the same simply by looking at glancing/crushing blows. The existence of a second type of "forced" hit is plausible, however. Perhaps some testing by lower level (weapon skill) characters against the boss could show whether it actually exists, and whether glancing blow % pushes it from the table (as would seem likely since I've never seen a lvl 1 *hit* the ?? dummy).

I do suggest that the concepts of this thread could be better shown in graphical form (stacked line graphs with y-axis error bars), showing in this case "apparent crit cap%|apparent "forced hit"% (ordinate) vs expertise (abscissa)" and "actual crit % vs paperdoll crit %" (showing the conclusion drawn from test3+test4)

Some in the Moonkin community have been getting 100% Starfire crits when crit capped for Lunar Eclipse. It hasn't to my knowledge been extensively tested but it would appear that for some abilities there is at present no crit depression being applied.

I've had parses with 100% Starfire crit, but not with enough spellcasts to reliably refute a 4.8% reduction. You could probably test this at a target dummy with a Mage giving you Imp. Scorch and Focus Magic. It's tricky to run a 100% Starfire crits, you need to rely on WiseEclipse mostly. But as long you get it most of the time, it would be easy enough to go for a few hundred casts and see if there are any non-crits while Eclipse is definitely up.

Spell crit reduction is conjectured to be smaller than the 4.8% number anyway - most suggestions I've heard are in the 2-3% range. Moreover, as spells are two-roll while melee attacks are one-roll, there's absolutely no reason to believe that the mechanics are the same anyway.

I recalled Vulajin or some Rogue did a long test that showed spell crit depression was 3%. Melee is 4.8%.

The "depression" you're talking about is just a sub-100% cap, right, not a flat crit% reduction?

That's sort of what we're here to figure out. For rogues, there is a straight percentage crit reduction against boss level mobs - that is, if you have less than 4.8% tooltip crit, your white attacks will never crit against an undebuffed target dummy. The assumption in the rogue community is that this is just a -4.8% modifier on your crit rate, and that's the end of that.

However, there is preliminary evidence that:
1) The size of the reduction may depend on class, and
2) The crit cap appears to not be in the place you'd expect based on a straight percentage reduction.

Hence, the point of this thread is to gather more data on the problem and see if we can figure out what's actually going on.

For warriors there seems to be a different reduction once naked (4.95% crit).
At 4.95% crit I got around 1.5% effective in around 10k hits. At 7.58% I had an effective crit of around 3% in 10k hits. However at 8.09% on 41194 swings I had 1371 crits for a 3.3% effective crit rate that seems to suggest the depression could be the same for both rogue and warriors, but something prevents it to go under a minimum crit value.

EDIT: For specials it seems possible to have 100% critrate overcapping the depression. I don't know if this value is the same for spells but for what I could test so far is higher than 4.33% and probably lower than 4.8%.

Tested with improved Overpower 2/2 (+50% crit) and 54.33% (49.33% + rampage). I underlined rampage refresh before the regular hit happened.



Test with 54.78% crit, no hits so far, but I should expect 1 in ~5k sample so I'll continue samplig. However this exclude the chance that there is a hard cap under 100% crit for specials.

From your data, I'd agree that the following all appear likely:

1) Warriors do receive a full 4.8% crit depression, but for some currently unknown reason the full effect doesn't occur at very low levels of crit. This can be seen in the disparity in the effect of crit depression between the test at 4.95% crit and 8.09% crit.

2) For white attacks, crit depression is essentially turning 4.8% worth of crits into hits rather than reducing your crit chance by 4.8% before building the table. This leaves an irremovable 4.8% hit chance on the table and effectively lowers what we considered the crit cap to be previously by 4.8% since it doesn't seem as if you can stack 4.8% more crit to overcome crit depression. This can be seen in all of your initial tests which are showing a reasonably near to 4.8% hit rate when you should be well beyond the crit cap.

3) For special physical attacks, crit depression is likely still 4.8% and the effects of crit depression can actually be overcome entirely, unlike with white attacks.

4.8% is for melee specials. The working theory I'd heard for spells was 3%, but I'm not sure that was ever conclusively proved. And it would seem to be refuted by your test anyway. Though it might differ depending on class.

Has there been any testing of this on the lvl80 dummy? ie, the 4.8% crit depression against bosses actually being a conversion of the 'first' 4.8% of crits into hits makes sense... But it could also simply be that the hit part of the combat table is a minimum of 5%. So when crit capped, do you get any hits on the lvl80 dummy? If not, the 4.8% crit->hit theory seems sound.

I have previously done tests of spell crit reduction for poisons and reported them here. I have observed poison crits with as low as 2.13% crit. The conclusion I made in that post is based on the assumption that spell crit depression was fixed. However, if spell crit reduction is lower at lower levels, it is possible that is starts off at around 2.1% and levels off at 3% at higher levels of crit.

In a separate testing I did almost 34k swings, however since I was testing mongoose and berserking ppm, I separated the results into subsets with and without mongoose up. The sample with mongoose had 8163 observations and 3563 crits, so 43.6% crit rate, while my theoretical crit with mongoose up was supposed to be 49.1%. Thus in those 8163 observations, crit depression was 5.46% (with 95% confidence interval of 4.96% to 5.95%). While mongoose was not up, crit depression was 4.92% on average with (4.66%;5.18%) confidence interval. I am inclined to think that instead of a fixed rate of crit depression, there is instead an increasing rate of crit depression, such as a fixed depression + a percentage of excess crit rate. For instance, for physical crits, the formula could be 3% base depression + 5% of (crit rate - 3%). For spell crit depression, it could be 2% + 5%*(crit rate - 2%).

Disregard, I completely forgot yellow attacks follow a 2 roll combat table instead of the white swing's 1.

By your data, seems that crit rating is affected by diminishing returns much like dodge or parry, and that the amount shown on the character panel is the value before the diminishing return is applied.

Could be that the innate 5% hit which can never be taken away on white hits can never crit?

Like a no-crit automatic hit which stays above the pve combat table ?

It's not 10% lower than thought, just 4.8%. The only difference is that since the 4.8% cannot be pushed off the table, that 4.8% is no longer part of the cap.

What am I missing then, I mean which 5% am I double counting?

- Was the crit cap not 100-D-M-G+5% under the crit depression hypothesis before the existence of Hellord's tests? Maybe we need numbers. Let's say D = 5%, M = 15%; we know G = 24, so 56% is left for crit and hit. If my crit rate with all buffs and debuffs is 61%, only 56% is used against bosses under the crit depression hypothesis. So I am at maximum of my crit, i.e. the crit cap is now 61%. Adding anything else is a waste for autoattacks.

- Is the crit cap not 100-D-M-G-5% now when we know 5% of crits are converted into hits? Using same numbers, 5% are dodges, 15% are misses, 24% are glances, and 5% are forced hits. Thus we have 100%-5%-15%-24%-5% = 51% left for crit and hit. So now adding any crit beyond 51% is not adding anything to my autoattack crits, so the new crit cap is 51%. Thus it seems to me there is a 10% difference between what we believed before and what we believe now.

I think when you say the difference in crit cap is only 5%, you are assuming there is both crit-to-hit conversion AND there is still a crit depression against bosses. While that may be true, I am advancing an alternative hypothesis that there never was a crit depression against bosses in first place, but instead there was always a conversion of 5% crits into hits when fighting against bosses. When we tested the crit depression before, we simply observed 5% less crits (and 5% more hits). That observation is consistent with both crit depression and hit conversion.

So in short, if the difference is not 10%, there should be some flawed logic in my calculation of crit caps before and after. Which one is it?

Before, the computation was the solution to the equation

(Crit-4.8) + Dodge + Miss + Glance = 100, so crit cap was 100-D-M-G+4.8, in your notation.

Under this theory, the equation is (Crit-4.8) + Dodge + Miss + Glance = 100 - 4.8 - that is, our crit is still reduced by 4.8, and we can't go up to 100% crit, only to 100-4.8 due to the forced misses. Rearranging to your notation, we get that the crit cap is simply 100-D-M-G, for a difference of 4.8. To look at it another way: when your "tooltip" crit, prior to crit depression, gets to 100-D-M-G, that indicates that all attacks, neglecting crit depression, should be Crits, dodges, misses, or glances; because of crit depression, 4.8% of attacks would instead be hits, but crit is just as capped.

That said: I'd hasten to point out that this is merely a theory right now - I don't think any comprehensive testing has been done to demonstrate that that's actually how it works. I just don't feel like we understand the problem well enough to comment definitively on what's going on.

So, per my previous post, I was not wholly confident in this theory, largely because warrior testing has indicated some manner of funkiness (particularly at low crit levels), which makes the mechanics there less than wholly understood. Hence, I opted to do some testing on a rogue, given that we have a decent handle on how rogue crit reduction seems to work - Vulajin's initial testing makes it pretty clear that it's a straight 4.8% reduction, even at low levels of crit. So, I hopped on PTR tonight (just in case there are any changes, and to save the cash I spent on un-speccing) and did some testing.

My gear setup had 20 expertise and 0 hit, and I specced out of all hit and expertise talents; thus, while dual-wielding, I have an expected miss rate of 27%, an expected glance rate of 24%, an expected dodge rate of 5.89%, and an expected parry rate of 13.39% (assuming the accepted values of 6.5% base dodge and 14% base parry). Thus, when attacking from the front, the available space for crit and miss is 29.8%. Under the old theory of crit reduction, I'd need at least 4.8% more than this to be capped, or 34.6%. And there is some uncertainty in the glance/parry rate numbers, so to be safe it would be good to have tooltip crit several percent above this. Well, in the test gear, my tooltip crit rate is 43.77%, a good 9% over the amount needed to cap even by the old theory. As such, if I exhibit any regular hits at all while attacking from the front, we know that there's some minimum "hit" rat - and indeed, in 6250 swings, I got 308 hits. Hence, the theory that there's a minimum hit rate seems pretty plausible.

However, there are multiple possible ways that such a minimum could be implemented - the next step is to test the conjecture that the minimum hit rate is precisely the 4.8% crit of crit depression we suffer; if this new theory is correct, we expect our hit table to be as follows:

Miss, 27%
Crit, 24.92%
Glance, 24%
Parry, 13.39%
Dodge, 5.89%
Hit, 4.8%

Using these, we can make a table of observed and expected attack counts for each result:

Result	Observed	Expected
Miss	1666	1687.5
Crit	1575	1557.5
Glance	1452	1500
Parry	878	836.9
Dodge	371	368.1
Hit	308	300

Clearly the agreement is pretty good - but *how* good? Well, fortunately there's an easy statistical test for this - we have 5 degrees of freedom, and a Chi-squared value of 4.263, which works out the a tail probability of 51.22%. Or, in English: assuming our theory is correct, we will get data with at least this much variance just over half the time, and less variance just under half the time. This is what's known in the business as "a stupidly good result". Basically, we couldn't ask for a better match to our theory than what we have. As such, while we certainly can't definitively prove our theory, it's definitely looking pretty good.

As such, the funkiness that warriors have been seeing notwithstanding, it's reasonably probable that rogues, at least, are having crit reduction applies as a forced conversion of 4.8% of our combat table from crits to hits, and that these 4.8% hits cannot be removed through any means - that is, the proposed theory is looking pretty good.

I do think there are still open questions in terms of what happens to warriors at low levels of crit (and similar effects) - but I'm at least feeling a bit more comfortable on the whole about this theory.

Don't you mean "available space for crit and hit is 29.8%"?


Ignore the italicized part, see edit.
The question is, I think, whether those 4.8% forced hits come from our crits or simply exist in the table the same way glancing blows do. In other words, if your crit rate had been exactly 24.92%, would you have observed 4.8% hits or 9.6%? This does matter because it would mean the crit cap in your example is effectively 24.92%, not 29.72% (or as Mavanas put it, 9.6% lower than previously thought).

If your testing answered that, I apologize, because I'm not seeing it.


EDIT: Having re-read Vulajin's testing, the conclusion was that he was not getting any crits with 4.8% crit chance on the paperdoll, so those forced hits do indeed come from our crits. Which means the crit cap is 4.8% lower than previously thought, not 9.6%. In your previous example, the crit cap would be 29.72%.

Okay, to clarify: yes, I meant crit and hit, not crit and miss. I'm not entirely sure what you're trying to say in the second part, so let me just review the logic to be clear what we think is happening.

Based on accepted values and testing, we have 27% miss, 24% glance, 13.39% parry, and 5.89% dodge. The total amount of hit table taken up by these options is 70.28%. Thus, there is 29.72% remaining for the last two outcomes, crit and hit. Thus, if no crit reduction existed, and there was no minimum on the number of hits we get, we'd expect all 29.72% of the table to be crits. Since tooltip crit rate is 43.77%, even if crit reduction were as high as 14.05%, we'd still expect to see no plain hits. Since we *do* see hits, we are forced to conclude that we *are* crit-capped, but are still seeing some hits anyway.

The obvious followup question is "how many" - well, the data (which matches hellord's warrior testing) is that there's around 5%. And we *know* that rogues experience an across-the-board crit reduction of 4.8% against boss level mobs. And it seems mighty suspicious that the number of hits we're seeing is very very close to our crit reduction.

Thus, the conclusion we draw is that crit reduction is implemented by a forced conversion of 4.8% crits to hits, which cannot be removed. Or, phrased alternately: the crit reduction is applied *after* crit capping is considered - which is perhaps the simplest way of seeing what the numbers should be. The logic then looks as follows: we've used up 70.28% on glance, miss, dodge, and parry. Thus, our crit is capped at 29.72%, and we have enough crit to hit that. And then, after our crit has been reduced from 43.77 to 29.72 due to crit cap, then and only then does crit reduction kick in and convert 4.8% crits to 4.8% hits, leaving 24.92% crit and 4.8% hit - which gives us exactly the hit table listed above.